Artificial intelligence in automatic passenger counting: cost-efficient validation using the partitioned equivalence test

Abstract

In Automatic Passenger Counting (APC), the demand for very low errors has been fueled by applications like revenue sharing, which amounts to massive annual sums. As a consequence, APC validation costs are increasing and this work presents a solution to increase the efficiency of initial or recurrent, e.g. yearly, validations. The new approach, the partitioned equivalence test, guarantees the same bounded, low user risk while reducing efforts in comparison to established (equivalence) test procedures. This involves a pre-classification step, which selects the more informative bits of data (e.g. footage). Different use cases are evaluated: from entirely manual to algorithmic, artificial intelligence assisted workflows. Already for manual counts, the new approach can be used as a drop-in replacement to reduce validation costs, while savings in algorithmic use cases demonstrated that cost halving is possible. Due to the user risk being controlled no additional technical requirements are introduced.